OBJECTIVES: We estimated the transition probabilities between different health states of esophageal cancer for the natural history of esophageal cancer. METHODS: We constructed a 30-cycle Markov model of esophageal cancer for Chinese women aged 40-69 years old.The model was initially populated with parameters(initial proportion of each health state,death probabilities,and transition probabilities) generated from screening programs,literature,and expert consultation.We adjusted transition probabilities until predicted prevalence of each health state and age-specific incidence of esophageal cancer were similar to findings from esophageal cancer screening in high-risk areas of China. RESULTS: Annual transition probabilities were 0.024,0.05,and 0.12 for normal to mild dysplasia,mild dysplasia to moderate dysplasia,and moderate dysplasia to severe dysplasia/carcinoma in situ (CIS),respectively.Age-specific progression probabilities were 0.08-0.18 for severe dysplasia to intramucosal carcinoma,0.40-0.87 for intramucosal carcinoma to submucosal carcinoma (T1N0M0),and 0.2-0.85 for submucosal carcinoma to invasive carcinoma.As for regression,transition probabilities were 0.05 for mild dysplasia to normal,and 0.08 for moderate dysplasia to mild dysplasia,and 0.09-0.17 for severe dysplasia/CIS to moderate dysplasia.Predicted incidence of esophageal cancer increased with age,and model generated estimates for prevalence and incidence were consistent with empirical data. CONCLUSIONS: We obtain reliable transition probabilities from a Markov model based on empirical data.Our model can be potentially useful for understanding the natural history of esophageal cancer.